2008 Society of Vacuum Coaters 505/856-7188 511

51st Annual Technical Conference Proceedings, Chicago, IL, April 1924, 2008 ISSN 0737-5921
ABSTRACT
Thin absorbing films are becoming common in optical coating
applications. Spectroscopic ellipsometry characterization of
such films requires proper techniques to insure unique results
for both thickness and optical constants.
This paper provides a review of ellipsometry methods to
characterize thin absorbing layers. All methods benefit from
either reducing unknown sample properties or increasing
measurement information. While both thickness and optical
constants for thin absorbing layers can be determined, mea-
surement sensitivity depends on method and implementation.
The advantages and limitations of each method are described,
along with examples from the optical coatings field.
INTRODUCTION
Spectroscopic Ellipsometry (SE) is routinely used to mea-
sure thickness and refractive index of transparent thin films.
Absorbing films are more difficult to characterize because
the optical constants (n and k) are often correlated with film
thickness. This calls into question the uniqueness of the
results. When properly implemented, SE measurements can
simultaneously and uniquely determine thin absorbing layer
thickness and optical constants.
Optical measurements of absorbing films are increasingly
common for two reasons. First, many applications now include
thin absorbing coatings as a critical element of the optical
design. For example, a thin Ag layer is added for low-emis-
sivity (low-e) coatings. In electrochromic layers, the exchange
of ions can vary the absorption within a layer, changing the
optical constants with applied voltage.
Second, many coatings are transparent at their design wave-
lengths, but absorbing in other spectral regions. It may be
beneficial to measure the coating at absorbing wavelengths
to better understand material properties or monitor coating
quality and performance. For example, transparent conductive
oxides are intentionally transparent at visible wavelengths, but
absorb longer infrared light. The optical response at infrared
wavelengths is a measure of the film conductivity. Figure 1
shows the variation in extinction coefficient for an indium
Spectroscopic Ellipsometry Methods for Thin Absorbing Coatings
J.N. Hilfiker and R.A. Synowicki, J.A. Woollam Company, Lincoln, NE;
and H.G. Tompkins, Consultant, Chandler, AZ
tin oxide (ITO) coating at various stages of processing. The
increase in near infrared absorption (larger extinction coef-
ficient) indicates improved conductivity.
Figure 1: ITO optical constants measured during processing.
Increased absorption at long wavelengths is due to free-carriers,
indicating improved conductivity from the annealing step.
SPECTROSCOPIC ELLIPSOMETRY
Spectroscopic Ellipsometry (SE) is an optical measurement
technique used to characterize film properties based on a
change in polarization as light interacts with layered materi-
als. SE theory is discussed elsewhere [1,2], but a few points
are worth consideration in the context of this work.
Ellipsometry measures the change in polarization caused by
interaction with thin coatings and substrates. This measurement
is recorded as two values related to the reflectance ratio of
p- and s- polarized light, given as an amplitude ratio, tan()
and phase difference, :
(1)
512
The polarization change caused by the sample is not of
primary interest. Rather, coating properties such as film
thickness and optical constants are desired. These properties
cause the resulting polarization change measured by SE.
Thus, ellipsometry is subject to the inverse problem. The
result is measured but the cause must be determined. This
is also true of other optical measurements. In this case, the
cause is the thickness and optical constants of each layer
in the optical coating stack, which is described by a model to
allow calculation of the optical response. Through regression
analysis, the unknown coating properties are found that best
generate a theoretical response to match the experimental
curves. This process is shown in Figure 2 and is referred to
as fitting the data.
ABSORBING FILM CHALLENGE
SE measurements of absorbing films are not as common
as measurement of transparent films due to the increased
complexity of the measurements and models. The optical
constants for absorbing layers contain two unknown values
per wavelength: index of refraction and extinction coefficient
(n and k). The extra level of sample complexity can lead to
correlation between parameters. To better understand this
situation, first consider the case of a transparent film.
Transparent Films: single-wavelength
Transparent films are commonly measured with spectroscopic
ellipsometry to determine film thickness and index. Consider
first a single-wavelength ellipsometry measurement. For a
transparent coating on known substrate, there are two un-
known sample properties: film thickness and index (n). The
ellipsometer measures two values, and . This provides
adequate information to solve for both sample properties.
However, the single-wavelength ellipsometric data is a peri-
odic function of film thickness so that different thicknesses
cycle through the same , values. Figure 3a shows the data
cycle as thickness increases. This leads to uncertainty in
the real thickness as any measured point along the circle can
correspond to multiple possible thicknesses separated by the
cycle-period.
Transparent Films: SE
To solve the uncertainty associated with single-wavelength
ellipsometry, the measurement is extended to cover multiple
angles and/or wavelengths. For example, SE data from a thin
transparent coating on known substrate would lead to +1
unknown sample properties ( index values plus 1 thickness)
where is the number of measured wavelengths. The number
of measured values is 2, as and are determined for each
wavelength. Thus, the problem becomes over-determined
when multiple wavelengths are measured.
The wavelength variation eliminates any misinterpretation
from thicknesses possible at different multiples of the periodic
cycle. The first three thicknesses from the cycle of Figure 3a
are shown versus wavelength in Figure 3b. Although the data
are identical at one wavelength, their wavelength dependence
is significantly different, ensuring a unique result.
Absorbing Films
For an absorbing film, the unknown sample properties increase
to 2+1: n(), k() and film thickness. Only 2 data values
(, ) are measured from a spectroscopic ellipsometer at one
angle. The initial reaction is to simply measure additional
data at a second angle of incidence, increasing the number
of measured values to 4. However, this will only succeed if
the second angle provides new information. Unfortunately for
many thin absorbing films, multiple angles provide essentially
the same information.
Figure 2: Flow chart for the ellipsometry process. Through
regression analysis, a model is adjusted to find the optical constants
and layer thicknesses that generate data curves that best match the
experimental data curves.
The final model is a layered description that best predicts
the measured data. However, the model must be assessed to
insure sensitivity and uniqueness.
The Mean Squared Error (MSE) is used to quantify the dif-
ference between a given model and the experimental data
[3]. A low MSE implies a good match between model and
experiment. However, a low MSE can also be obtained if too
many sample parameters are fit. This leads to a non-unique
result. To ensure the final model is correct, it is important to
obtain adequate data content to determine all sample unknowns
using a minimum number of free fit parameters.
513
Figure 3: (a) Data cycle versus thickness for a TiO
2
coating on
fused silica. The data at a single-wavelength are equivalent for
200nm, 400nm, 600nm etc. (b) Spectroscopic data for the same three
coatings at 70 angle show the match for data at 633nm, but have
very different spectral responses, which allows SE measurements to
uniquely determine between the possible thicknesses from cycle.
Consider a thin TiN film deposited on Si substrate. SE data
are collected at 4 angles of incidence, providing 8 total
data points. This should be adequate to solve 2+1 unknown
sample properties, unless the different angles provide es-
sentially the same information. However, this case remains
under-determined, with multiple solution sets of thickness
and optical constants that provide the same data fit quality
(similar low MSE value). Figure 4 shows the uniqueness test
using data from all four angles for the thin TiN layer on Si.
There are multiple thicknesses that provide the same MSE
accomplished by fitting different optical constants for each
thickness. Which answer is correct? Because all combina-
tions fit the data with similar resulting MSE, a unique result
is not obtained, even though plenty of experimental data
had apparently been collected. This is the primary problem
with absorbing films. How to ensure a unique solution? It is
important to consider the information content of the data
rather than just the amount of measured data.
Figure 4: Example of correlation, where multiple sets of thickness
and n,k values provide similar quality fit (low MSE). Thus, it is not
possible to determine which result is correct. More information is
needed.
CHARACTERIZING ABSORBING FILMS:
METHOD REVIEW
To insure a unique analysis of absorbing films, it is important
to apply special measurement and analysis procedures. These
methods have been discussed when applied to Cr metal coat-
ings [4], but are reviewed here with consideration for optical
coatings. Each method is described along with its inherent
advantages and limitations. All methods either increase the
measured data content or reduce the number of unknown
sample properties. Choosing the correct method can make the
difference between an under-determined, non-unique solution
and accurately determining the optical coating properties.
Opaque Coatings
The most common method applied to absorbing films is to
deposit a thick, opaque layer of the same material. Because
the coating is opaque, no light can penetrate the film and the
thickness is no longer a consideration. This reduces the num-
ber of unknown sample properties (no thickness to consider)
and allows a direct inversion of () and () to determine
the n() and k() for the film [3]. These optical constants are
referred to as pseudo values because the direct inversion
assumes a sharp interface between the absorbing coating and
ambient. Thus, any surface roughness or oxidation is ignored
and may affect the accuracy of the measured result.
This opaque film technique can provide a basic description
of the optical response from a specific material. However,
the resulting optical constants may not represent the values
from a thinner layer for two reasons. First, as mentioned
the direct inversion ignores all effects from surface layers
514
(oxides, roughness, etc.) which may change from thin to thick
layers. Second, for many materials, the optical properties
will vary with film thickness, especially in the sub-100nm
regime under consideration. Figure 5 shows optical constants
determined from a series of Ag films with varying thickness.
Note the changes in optical properties as thickness increases.
For this case, the opaque coating does not describe the optical
constants of thinner layers.
the optical constants of a material versus wavelength. This
significantly reduces the total number of variable fit param-
eters in the model typically 5 to 25 parameters are varied
to match data from hundreds of wavelengths. However, the
dispersion equation does not add information content to
the experimental data, or significantly reduce the amount
of information unknown about the film. The same coating
properties (thickness, n() and k()) need to be determined.
The dispersion equation only reduces the number of fit pa-
rameters and constrains the possible choices that are allowed
during the fit. The key advantage of a dispersion equation is
that many are developed according to physical descriptions
of material properties, relating changes in k to changes in n
and vice-versa. As such, many functions have been developed
over the years which serve as dispersion equations and en-
force Kramers-Kronig consistency. This ensures the resulting
optical constants retain a physically reasonable shape that is
possible in nature. Thus, dispersion equations can eliminate
many of the unphysical possibilities of a direct wavelength-
by-wavelength fit to optical constants.
One limitation of optical constant parameterization is that
the final dispersion equation must match the shape of the
materials real optical constants. This can be easy for amor-
phous semiconductors such as amorphous silicon, which have
broad features and can be described with a single Tauc-Lorentz
or Cody-Lorentz oscillator [5]. For example, Figure 6 shows
the imaginary dielectric function for a series of amorphous,
microcrystalline, and polycrystalline silicon films used for
photovoltaic applications. A Kramers-Kronig consistent dis-
persion equation with five to ten free parameters can describe
the optical constants over the full spectral range.
Figure 5: Optical constants for Ag thin films versus thickness.
Transparent Wavelength Region
Many coatings are absorbing at some wavelengths, but
transparent at others. While metals generally dont meet this
criterion, almost all dielectrics, organics, and semiconductors
will. A common and successful approach for these coatings
involves two steps. First, the thickness is determined from
the transparent region, where only thickness and n() are
unknown. After the thickness is accurately determined, n()
and k() can be determined in the absorbing spectral region
on a wavelength-by-wavelength basis. Here, again the number
of unknown sample properties are reduced, as thickness was
determined from the transparent region and is fixed.
The main limitation of this method is the requirement of
film transparency over some spectral range accessible by the
measurement tool. As mentioned earlier, metals absorb at all
wavelengths. Low bandgap semiconductors absorb light from
the ultraviolet to the near infrared and only become transpar-
ent at mid-infrared wavelengths. While IR ellipsometers are
available, they are less common.
Optical Constant Parameterization
Another common method for absorbing film characterization
is the use of a mathematical dispersion equation to describe
Figure 6: Imaginary dielectric functions developed using dispersion
equations for amorphous, micro-crystalline, and poly-crystalline
silicon films.
Optical constant parameterization is more difficult for organics,
crystalline semiconductors, and metals that exhibit multiple
absorption features. Here, a summation of oscillator terms are
combined to describe the optical constants versus wavelength.
As the dispersion equation becomes more complex, it also
515
becomes less effective at reducing correlation. Thus, optical
constant parameterization is often best utilized in combination
with the other methods described in this review.
Interference Enhancement
While multiple angles do not necessarily provide additional
information for all samples, there is a special type of sample
structure that benefits from multiple angles. If the absorbing
film is deposited over a thick dielectric layer or stack, the
interaction between light and the absorbing film is signifi-
cantly modified by the underlying coating(s). This is referred
to as interference enhancement and was first demonstrated by
McGahan et al. for amorphous carbon films [6].
To demonstrate interference enhancement, compare the
uniqueness tests of two TiN coatings. The first is deposited
directly on a Si substrate and is shown in Figure 4. Although
SE data are collected at four different angles of incidence, each
of the multiple angles provide essentially the same informa-
tion and there is correlation between thickness and optical
constants. A similar TiN film is also deposited over a thick
SiO
2
coating on Si. The underlying dielectric enhances the
light interaction with the absorbing film at multiple angles.
Because there is significant change versus angle, new infor-
mation is measured at each angle and a unique solution for
optical constants and thickness is obtained. In Figure 7 this
is shown by a Uniqueness test that only has a single unique
result (MSE minimum) for optical constants and thickness.
measurement techniques. This has been demonstrated as early
as 1992 for diamond-like carbon (DLC) films and has been
commonly used for absorbing coatings on glass and plastic
substrates ever since [7]. The extra intensity measurement
helps break the correlation between thickness and optical
constants and allows a unique result.
Consider the uniqueness test results shown in Figure 8. When
only SE data are considered, the fit result is not unique. Add-
ing transmission intensity to the fit provides the additional
information to break correlation between optical constants
and thickness and provides a single unique solution (MSE
minimum) for the coating optical constants and thickness.
Figure 7: Uniqueness test for a thin absorbing TiN coating deposited
on Si substrate with thick SiO
2
dielectric layer. The underlying
dielectric enhances the information content available from multiple
angle measurements, which provides a unique solution for optical
constants and thickness.
Simultaneous SE and Intensity
Another method that increases the measured information con-
tent is to combine ellipsometry measurements with intensity
reflectance or transmittance measurements from the same
coating. The most common implementation involves transmit-
tance, as it is more sensitive to the absorption than reflected
Figure 8: Uniqueness test shown for Cr thin film, comparing results
when only SE data are analyzed and when SE and Transmission are
simultaneously analyzed.
Multiple Sample Analysis
Another method to increase measured data content is to mea-
sure multiple samples of the same coating with different film
thickness. The data are fit simultaneously using a common set
of optical constants. Consider measurement of three different
samples that contain an absorbing coating, but with different
film thickness. The unknown sample properties include three
thickness values, but only one set of optical constants: n()
and k(). The measured , data include 2 values from each
sample, which totals 6 for three samples. The information
content will be different from each sample as long as the
light interaction is different (different path length through
each film). Thus, there are 6 values to uniquely determine
the 2+3 unknown sample properties.
In-Situ Monitoring
In-situ SE measurements collect data during the deposition
process. This provides access to the same coating at different
points during deposition (different film thickness). This is
equivalent to the benefits from multiple sample analysis, but
applied to the same coating and often includes tens or hundreds
of different thicknesses. As all measurements are from the
516
same coating the optical constants are often more consistent
than with the multi-sample approach. This method is the most
successful approach of all the methods surveyed. However, it
requires integration of the ellipsometer on to the deposition
process chamber. As an added benefit of integrating SE tool
onto the process chamber, the real-time monitoring of films
allows control of the final film thickness [8].
Multiple Ambient Method
The final method to collect additional data from a sample
involves varying the ambient surrounding the sample (e.g. air
and water). If the ambient index changes, the light interaction
with the coating also changes and provides new informa-
tion. Thus, measurements with multiple ambients should be
adequate to solve for thickness and optical constants of an
absorbing film. The main limitation to this technique is con-
sideration of how the coating may change in different ambient.
Some coatings may absorb some of the ambient, which will
modify the coating optical constants. This will increase the
complexity of the data analysis and reduce the effectiveness
of determining a unique result for absorbing films. However,
this is used to an advantage for porous coatings to help study
the pore size.
COMBINED METHODS
In practice, the methods described to help measure absorbing
films are best used in combination. This provides the benefits
from each method and overcomes many of their individual
limitations. A perfect example of combined methods is dem-
onstrated by Pribil et al [9]. They combined three of the above
methods in their study of thin absorbing metal layers. Both SE
and Transmittance were measured in-situ during the deposi-
tion of the metals. During data analysis, the optical constants
were parameterized with a dispersion equation. Thus, three
methods were combined to ensure unique results.
SUMMARY
A variety of methods have been presented that allow spectro-
scopic ellipsometry measurements of thin absorbing films.
Each method helps reduce the correlation between thickness
and optical constants (n and k) to ensure a unique result.
Many methods succeed by reducing the number of free-pa-
rameters in the model analysis. These include measurement
of an opaque absorbing film, measuring a transparent spectral
region to fix the thickness before characterizing the absorbing
region, and optical constant parameterization. Other meth-
ods rely on collecting additional data from the sample that
provides additional information to be used in the fit. These
methods include multiple angle measurements that benefit
from interference enhancement, combined SE and intensity
measurements, multiple sample analysis, in-situ monitoring,
and multiple ambient measurements. All methods work well
on a variety of thin absorbing and semi-absorbing film, but
have their own limitations which need to be considered when
characterizing such coatings. Best results often result from
the combination of multiple available methods.
REFERENCES
1. H.G. Tompkins, E.A. Irene (Eds.), Handbook of El-
lipsometry, William Andrew Publishing, Norwich NY,
2005.
2. H. Fujiwara, Spectroscopic Ellipsometry Principles and
Applications, J ohn Wiley & Sons Ltd, West Sussex,
England, 2007.
3. J . Hilfiker et al. Survey of methods to characterize thin
absorbing films with Spectroscopic Ellipsometry, Thin
Solid Films, 2008, In Press.
4. B. J ohs, J .A. Woollam, C.M. Herzinger, J . Hilfiker,
R. Synowicki, C. Bungay, in: G.A. Al-J umaily (Ed.),
Overview of Variable Angle Spectroscopic Ellipsometry
Part I, Proceedings of SPIE, Denver, U.S.A., July 18-19,
1999, Critical Review of Optical Science And Technology
CR72 1999.
5. R.W. Collins and A.S. Ferlauto, in: H.G. Tompkins, E.A.
Irene (Eds), Handbook of Ellipsometry, William Andrew
Publishing, Norwich NY, pp. 93-235, 2005.
6. W.A. McGahan, B. J ohs, J .A. Woollam, Techniques for
ellipsometric measurement of the thickness and optical
constants of thin absorbing films, Thin Solid Films 234
pp. 443-446, 1993.
7. B. J ohs, D. Meyer, J .A.Woollam, J .F. Elman, T.E. Long,
R.F. Edgerton, J .T. Koberstein, Characterization of In-
homogeneous and Absorbing Thin Films by Combined
Spectroscopic Ellipsometry, Reflection, and Transmission
Measurements, Optical Interference Coatings Technical
Digest 15 pp. 443-436, 1992.
8. B. J ohs et al. Recent developments in spectroscopic
ellipsometry for in situ applications SPIE Proc. 4449
pp. 41-57, 2001.
9. G. Pribil, B. J ohs, N.J . Ianno, Dielectric function of thin
metal films by combined in situ transmission ellipsometry
and intensity measurements Thin Solid Films 455-456
pp. 443-449, 2004.